[clang-format] Fix a bug in annotating CastRParen

[owenca]

Fixes #102102.

[llvmbot]

@llvm/pr-subscribers-clang-format

Author: Owen Pan (owenca)

Changes

Fixes #102102.

---

Full diff: https://github.com/llvm/llvm-project/pull/102261.diff

2 Files Affected:

• (modified) clang/lib/Format/TokenAnnotator.cpp (+10-1)
• (modified) clang/unittests/Format/TokenAnnotatorTest.cpp (+7)

diff --git a/clang/lib/Format/TokenAnnotator.cpp b/clang/lib/Format/TokenAnnotator.cpp
index 4ed3e9d0e8e855..e1c9f37c6e8dc2 100644
--- a/clang/lib/Format/TokenAnnotator.cpp
+++ b/clang/lib/Format/TokenAnnotator.cpp
@@ -2874,9 +2874,18 @@ class AnnotatingParser {
       return false;
 
     // Search for unexpected tokens.
-    for (auto *Prev = BeforeRParen; Prev != LParen; Prev = Prev->Previous)
+    for (auto *Prev = BeforeRParen; Prev != LParen; Prev = Prev->Previous) {
+      if (Prev->is(tok::r_paren)) {
+        Prev = Prev->MatchingParen;
+        if (!Prev)
+          return false;
+        if (Prev->is(TT_FunctionTypeLParen))
+          break;
+        continue;
+      }
       if (!Prev->isOneOf(tok::kw_const, tok::identifier, tok::coloncolon))
         return false;
+    }
 
     return true;
   }
diff --git a/clang/unittests/Format/TokenAnnotatorTest.cpp b/clang/unittests/Format/TokenAnnotatorTest.cpp
index 6b71d0d18cf7a4..cbbe550a79ebc2 100644
--- a/clang/unittests/Format/TokenAnnotatorTest.cpp
+++ b/clang/unittests/Format/TokenAnnotatorTest.cpp
@@ -740,6 +740,13 @@ TEST_F(TokenAnnotatorTest, UnderstandsCasts) {
   EXPECT_TOKEN(Tokens[10], tok::r_paren, TT_Unknown);
   EXPECT_TOKEN(Tokens[11], tok::amp, TT_BinaryOperator);
 
+  Tokens = annotate("func((void (*)())&a);");
+  ASSERT_EQ(Tokens.size(), 15u) << Tokens;
+  EXPECT_TOKEN(Tokens[4], tok::l_paren, TT_FunctionTypeLParen);
+  EXPECT_TOKEN(Tokens[5], tok::star, TT_PointerOrReference);
+  EXPECT_TOKEN(Tokens[9], tok::r_paren, TT_CastRParen);
+  EXPECT_TOKEN(Tokens[10], tok::amp, TT_UnaryOperator);
+
   auto Style = getLLVMStyle();
   Style.TypeNames.push_back("Foo");
   Tokens = annotate("#define FOO(bar) foo((Foo)&bar)", Style);

[prj-]

I'm sorry I'm not familiar with LLVM release cycle, but is it too late for this to make it into the release/19.x branch?

[owenca]

I'm sorry I'm not familiar with LLVM release cycle, but is it too late for this to make it into the release/19.x branch?

It's not too late, but in general we backport bug fixes only if the bugs are labeled as invalid-code-generation, crash-on-valid, or regression. In this particular case though, the fix is localized and fairly simple, and has been approved by all clang-format maintainers, so I've decided to backport it to 19.x per your request.

[owenca]

/cherry-pick 8c7a038

[llvmbot]

/pull-request #102419

[prj-]

Thank you very much.

[prj-]

This introduced a regression though.

-      for (i = 0, k = 1; i < ((PetscInt)local_n0) * partial_dim; i++, k++) {
+      for (i = 0, k = 1; i < ((PetscInt)local_n0)*partial_dim; i++, k++) {

[prj-]

Nevermind, this bug (not sure if it's a bug or not) is already present in clang-format 18. See #102727.

[krasimirgg]

It appears that this introduced a regression in code like this:

int d = 0;
int c(int i) { return i; }
void a() { double b = ((double)c(0)) * d / (double)0; }

After this patch, the last line gets formatted as:

void a() { double b = ((double)c(0))*d / (double)0; }

[prj-]

What's your version?

$ clang-format-20 gh.txt
void a() { double b = ((double)c(0)) * d / (double)0; }

[krasimirgg]

Ohh sorry, I had started with a stale clang-format commit. This is a non-issue at HEAD.

[kadircet]

there still seems to be a regression when the operator is -, filed #107568